Automatic Modulation Recognition for Overlapping Radar Signals based on Multi-Domain SE-ResNeXt

In the field of electronic reconnaissance, due to the complex electromagnetic environment, the receiver receives the high density signals, and two or even more radar signals are often overlapped. However, the features of the signals are completely interleaved. Especially under low signal-to-noise ratio (SNR), the signals become weak and messy, and the automatic modulation recognition is quite difficult. Through Fourier transform, the overlapping signals can be separated in frequency domain to some extent. The Autocorrelation Function (ACF) has some effect on de-noising, and the time-frequency analysis can reflect the instantaneous frequency of the signal. Therefore, this paper proposes a Multi-Domain Squeeze-and-Excitation ResNeXt (SE-ResNeXt) based on multi-branch feature fusion. In addition to the time domain, the signal is transformed to frequency domain, autocorrelation domain and time-frequency domain, and the high-dimensional feature vectors of signals in different domains are extracted by SE-ResNeXt. Finally, the Squeeze-and-Excitation (SE) block fuses the high-dimensional features of the four branches, the automatic modulation recognition of the overlapping radar signal is realized. The simulation results of six typical radar signals show that the algorithm is robust and effective. At -4dB, the recognition accuracy of the proposed method is more than 92%. Compared with other methods, the proposed method can achieve better performance under low SNR.